基于YOLOv5s改进的口罩佩戴检测算法  

Based on the improved YOLOv5s mask wearing detection algorithm

作　　者：葛延良[1] 李德鑫 王冬梅[1] 董太极 贺敏 GE Yanliang;LI Dexin;WANG Dongmei;DONG Taiji;HE Min(School of Electrical&Information Engineering,Northeast Petroleum University,Daqing 163000,China;Daqing Branch,China Mobile Group Heilongjiang Limited Company,Daqing 163000,China)

机构地区：[1]东北石油大学电气信息工程学院,大庆163000 [2]中国移动通信集团黑龙江有限公司大庆分公司,大庆163000

出　　处：《黑龙江大学自然科学学报》2023年第3期362-368,共7页Journal of Natural Science of Heilongjiang University

基　　金：黑龙江省自然科学基金(LH2020F005)。

摘　　要：由于新型冠状病毒肺炎的爆发,口罩成为人们日常生活中必需品。为了识别与检测人们是否佩戴口罩,提出了一种基于改进的YOLOv5s口罩佩戴检测算法。通过在YOLOv5s主干网络引入改进的自适应的协调注意力机制模块(Coordinate attention-activate or not,CA-A)提升网络的特征提取能力,解决了错误检测和漏检的问题。以新的损失函数AD-CIoU代替CIoU损失函数,作为回归损失函数,提升了边界框的定位精确度。实验表明,与原始模型算法相比,所提出的模型算法平均精度mAP值达到96.1%,提升了1.7%,具有较好的检测精度,可以满足目标检测应用需求。Due to the outbreak of novel coronavirus pneumonia,the mask has become a necessity of people’s daily life.In order to identify and detect whether people wear masks or not,the improved YOLOv5s mask wearing detection algorithm is proposed based on YOLOv5s.By introducing an improved attention mechanism Coordinate attention-activate or not(CA-A)module in the YOLOv5s backbone network,the feature extraction capability of this network is improved,and the problem of false detection and missed detection is solved.The new loss function AD-CIoU is proposed to replace the CIoU loss function as the regression loss function to achieve the purpose of improving the localization accuracy of the bounding box.The experimental results show that the average accuracy mAP value of the proposed model algorithm reaches 96.1%,which is 1.7%higher than the original model algorithm,and has a better detection accuracy,which can meet the practical application requirements of target detection.

关 键 词：计算机视觉 YOLOv5s 口罩佩戴检测 CA-A注意力 AD-CIoU损失函数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]